Improving Pipeline Opportunity Assesment Using Predictive Analytics

Collaborative for CUBES

16 November, 2021

Background

Goal
- To increase profitability through selective bidding of opportunities
Process
- Use historical data to understand drivers of bid success
- Showcase that a predictive model performs well with available data
- Deploy predictive model to pick opportunities
Results
- Ensemble model is able to predict the bid winner with 98.1% accuracy in the training data (data used to build the model)
- Ensemble model is able to predict the bid winner with 76% accuracy in the test data (data unseen by the model)
Deployment
- Developed a web application of the predictive model to be used for predictig win probability

Data

Data available on all bids Sodexo quoted since 2010
Number of bids

## [1] 3746

Distribution of opportunities over Fiscal Years

Data

Distribution of Opportunities by Old Provider

Data

Distribution of Opportunities by New Provider

Old Provider vs New Provider

Number of Opportunities

Old Provider vs New Provider

Revenue - Millions of Euros

Old Provider vs New Provider - Sodexo

Old Provider vs New Provider - By Opportunity Global Region

Old Provider vs New Provider - By Service Level 1

Old Provider vs New Provider - By Strategic Segment L1

Old Provider vs New Provider - By Regional Strategic Account

Regression Analysis - Logistic Regression

Regression Results

**Logistic Regression Model Predicting Sodexo Winning the Bid**

	Sodexo Winning the Bid
	Full Data

`Old Provider`Aramark	-2.130^***
	(0.329)

`Old Provider`Compass	-1.384^***
	(0.165)

`Old Provider`Other	-1.163^***
	(0.091)

`Opportunity Global Region`AMENA	-13.105
	(309.686)

`Opportunity Global Region`ASIA PACIFIC	0.267^**
	(0.125)

`Opportunity Global Region`BRAZIL	-0.510^***
	(0.140)

`Opportunity Global Region`CONTINENTAL EUROPE	0.588^***
	(0.181)

`Opportunity Global Region`LATAM	-0.146
	(0.142)

`Opportunity Global Region`NO REGION	-2.597^**
	(1.021)

`Opportunity Global Region`UK	I
	(0.149)

`Service Level 1`Food	0.062
	(0.145)

`Service Level 1`Integrated services	-0.195
	(0.142)

`Service Level 1`Soft FM	0.291^**
	(0.140)

`Strategic Segment L1`Corporate Services	0.347
	(0.493)

`Strategic Segment L1`Energy	-0.086
	(0.100)

`Strategic Segment L1`Government Services	0.508
	(0.603)

`Strategic Segment L1`Healthcare	-14.210
	(882.743)

`Strategic Segment L1`No Segment	-13.030
	(509.088)

`Strategic Segment L1`Schools	15.173
	(882.743)

`Regional Strategic Account`1	0.267
	(0.209)

year_greater_than_20161	0.538^***
	(0.136)

`Revenue (Annual) (converted) in Mill Euros`	-0.051^***
	(0.016)

`Margin (GP/BR)`	-0.006^***
	(0.002)

Constant	-0.226
	(0.201)

N	3,746
Log Likelihood	-2,066.078
Akaike Inf. Crit.	4,180.155

Notes:	^***Significant at the 1 percent level.
	^**Significant at the 5 percent level.
	^*Significant at the 10 percent level.
	Base Case - Old Provider = ‘Sodexo’; Opportunity Global Regian = ‘NORAM’; Service Level 1 = ‘Hard FM’; Strategic Segment L1 =‘Mining’

Regression Analysis - Multinomial Logistic Regression

Model Accuracy

Regression Results

**Multinomial Logistic Regression Model Predicting New Provider Relative to Sodexo**

	Aramark	Compass	Other

`Old Provider`Aramark	3.970^***	0.600	2.050^***
	(0.519)	(0.503)	(0.347)

`Old Provider`Compass	1.417^**	1.468^***	1.338^***
	(0.560)	(0.205)	(0.179)

`Old Provider`Other	-0.112	-1.050^***	1.597^***
	(0.486)	(0.166)	(0.100)

`Opportunity Global Region`AMENA	4.584^***	-6.646^***	44.918^***
	(0.000)	(0.000)	(0.000)

`Opportunity Global Region`ASIA PACIFIC	-36.288^***	0.180	-0.334^***
	(0.000)	(0.236)	(0.129)

`Opportunity Global Region`BRAZIL	-32.785	0.446	0.509^***
		(0.274)	(0.143)

`Opportunity Global Region`CONTINENTAL EUROPE	-0.550	-0.234	-0.662^***
	(0.890)	(0.301)	(0.191)

`Opportunity Global Region`LATAM	0.793	-0.025	0.158
	(0.625)	(0.295)	(0.147)

`Opportunity Global Region`NO REGION	-5.981^***	-90.517	2.655^***
	(0.00004)		(1.024)

`Opportunity Global Region`UK	I	0.859	-0.654^**
	(0.587)	(0.291)	(0.156)

`Service Level 1`Food	0.549	0.184	-0.122
	(0.812)	(0.284)	(0.149)

`Service Level 1`Integrated services	0.841	0.581^**	0.119
	(0.872)	(0.289)	(0.146)

`Service Level 1`Soft FM	0.280	-0.258	-0.300^**
	(0.818)	(0.296)	(0.143)

`Strategic Segment L1`Corporate Services	-24.634^***	-52.070	-0.341
	(0.000)		(0.500)

`Strategic Segment L1`Energy	-1.052^**	0.439^**	0.055
	(0.527)	(0.199)	(0.103)

`Strategic Segment L1`Government Services	-33.784	-44.892^***	-0.474
		(0.000)	(0.616)

`Strategic Segment L1`Healthcare	0.605^***	-1.528^***	46.596^***
	(0.000)	(0.000)	(0.000)

`Strategic Segment L1`No Segment	-0.164^***	-5.814	44.883^***
	(0.000)		(0.000)

`Strategic Segment L1`Schools	-1.213^***	-21.328^***	-52.843^***
	(0.000)	(0.000)	(0.000)

`Regional Strategic Account`1	-18.609^***	0.026	-0.270
	(0.000)	(0.409)	(0.215)

year_greater_than_20161	-0.886	-1.213^***	-0.370^***
	(0.571)	(0.226)	(0.141)

`Revenue (Annual) (converted) in Mill Euros`	0.074^**	0.099^***	0.043^***
	(0.037)	(0.020)	(0.016)

`Margin (GP/BR)`	-0.007	0.008^***	0.005^***
	(0.010)	(0.003)	(0.002)

Constant	-2.951^***	-1.032^***	-0.290
	(0.987)	(0.370)	(0.210)

Akaike Inf. Crit.	6,008.598	6,008.598	6,008.598

Notes:	^***Significant at the 1 percent level.
	^**Significant at the 5 percent level.
	^*Significant at the 10 percent level.
	Base Case - Old Provider = ‘Sodexo’; Opportunity Global Regian = ‘NORAM’; Service Level 1 = ‘Hard FM’; Strategic Segment L1 =‘Mining’

Using Machine Learning to Improve Predictive Accuracy

Supervised Machine Learning Approach Called Ensemble Learning

Ensemble Learning

Model Stack
- Random Forest
- Gradient Boosted Machines
- Multinomial Logistic Regression
- Naive Bayes Classification
- Support Vector Machines

Meta Learner
- Random Forest

Ensemble Learning

To avoid over-fitting, we split the data into Train and Test samples
- 80% Train data and 20% Test data
Model Accuracy

Deployment

Link to the App to Predict Win Probability